Power Digital
Omega is an always-on analyst that scans every client every day, finds what's wrong and pushes a ready-to-act recommendation before anyone asks. It turns a senior strategist's judgment into a service that runs around the clock.
Ryne Benson
Omega · Power Digital · 2 min read
Power Digital runs two flagship AI products on one shared layer. Iris is the analyst you can interrupt: ask about a client and get a cited answer. Omega is the analyst that interrupts you. It watches every client against their own goals and tells the account team what to fix, often before the next client call.
I led the migration of Iris and Omega onto AWS, off Render, Supabase and Managed Agents and onto Cognito, CloudFront, DynamoDB and Bedrock AgentCore, with both products live throughout. I'm also on the team that owns the agent platform they run on, its MCP layer and the pipeline that ships every runtime.
How it works
A KPI moves against the goal
A daily scan checks each client against their own targets, like acquisition cost over goal or revenue down, across paid social, search, shopping, lifecycle, programmatic and affiliate.
Walk the tree to a root cause
Decision trees ask yes-or-no questions that narrow a broad problem to one specific, fixable cause, the way a senior strategist would think it through.
Recommend in plain language
Omega writes a card with what's wrong, what to do, who owns it and what the deliverable is.
One click to approve
A person approves the recommendation. That click is the only human step between the signal and the finished work.
A specialist delivers
A specialist agent builds the deliverable, an audience plan, media plan, creative brief or report, grades it until it passes and files it to the client's Drive, often the same morning.
Decisions worth explaining
Propose, never touch
Omega recommends but doesn't change live ad accounts. Everything runs in a walled environment with a human approval gate between diagnosis and action.
One tool layer for both products
Iris and Omega call the same governed catalog of client truth, exposed once as MCP tools: the knowledge graph, the account team, governed metrics, churn and expansion scores and call intelligence. The two products can't disagree on a number.
One quality bar
Omega's recommendations are graded against the standard Iris already set. Giving Omega Iris's framing and context took a real recommendation from a reject to a pass without changing Omega's engine.
It learns every night
A nightly pass consolidates each client's history into reusable playbooks, so recommendations sharpen over time and the knowledge stays when people move on.
A migration with no downtime
Moving Iris and Omega onto AWS meant cutting over auth, data and agent runtimes while both stayed in daily use.
By the numbers
- 8 channels covered
- ~480 diagnostic paths
- 1 click from signal to work
- 191 agent runtimes on the platform
The stack
- Interface. Next.js, React, Tailwind CSS, shadcn/ui, AWS IoT
- Agents. Claude Agent SDK, Bedrock AgentCore, Managed Agents, Decision trees
- Tools. MCP, FastMCP, AgentCore Gateway, Knowledge graph, Looker
- Data. DynamoDB, S3, SSM Parameter Store
- Platform. SST, Pulumi, Lambda, Cognito, CloudFront, SQS
- Delivery. GitHub Actions, Docker, semantic-release